On 09/04/2018 16:24, David Jones wrote:
It's a few months that I'm using rspamd. I wrote a dedicated plugin for
amavisd-new and I use it's scoring together with SA's.
Been playing around with rspamd over the weekend to see how it
compares and so far not that impressed. It has a few features that
are interesting like the MX check but other than that it's not as
impressive as the author makes it out to be on the website comparing
it to SA.
It claims to have better Bayes but so far I am seeing identical
results after identical training.
IMHO to reach satisfying results you have to train it a lot more that
SA, but in the long run it's a nice addition. My empirical observations
suggests that it gets better after at least 3000 ham and spam email
learned. It's also cool that you can train both global and per-domain
bayes, very useful if you have a multitenant installation with a lot of